24.962
Page 1
5/23/05
Sympathy (continued and ended)
1. Summary:
There are over and under-generation problems with Sympathy. One of the clearest one is the following,
involving undergeneration: Sympathy cannot express opaque orders among allophonic processes.
2. German (Ito and Mester 2002)
ç in complementary distribution with x
After [+back]
ç
x
Â
å8
dorsal fric -> [+back]/ [+back, -cons]_
After others1 and elsewhere
manche [manç´], solche [zolç´], Pech [peç],
ich [iç], möchte [mPçt´], Bücher [byç´r],
Charisma [çarisma], chtonisch [çtoniS]
Tuche [tux´], Bach [bax], doch [dox], auch
[aux], mochte [moxt´]
 in complementary distribution with å8
In onset
Rat [RAt], rot [Rot], Rätsel [REts´l]
3. Alternations
Scharr-en [SAR´n]
moch-t-e [moxt´]
[son, -lateral, -nas, -syll] -> å8 / in coda
In coda
wir [viå8], werden [vEå8d´n], Uhr [uå8]
Scharr
möch-t-e
[SAå8]
‘dig’
[mPçt´] ‘wish, want, like to’
These establish that the R and å8, on the one hand, and x vs. ç, on the other, are not only in complementary
distribution but also that one becomes the other, in contexts where it is impermissible. Simple complementary
distribution doesn’t tell us this.
4. Â -lenition counterfeeds-in-context ç-backing: first analyzed in rule terms, assuming /ç/ and /Â/
UR
ç-backing
 –lenition
fürchten [fyå8çt´n]
[fyRçt´n]
Bach [bax]
[baç]
[bax]
[fyå8çt´n]
Aside: Segments conditioning x-backing are [-front,-cons] (Ito & Mester 2002). Then [å8] resulting from
lenition should trigger the x-process, but for the opacity effect: it’s as non-front as [A] and a glide, thus [-cons].
But perhaps x-triggers must be [+syllabic], or, alternatively, perhaps [å8] is central while x-triggers have to be
[+back]: then we should not expect lenited R to trigger backing. In that case the interaction is non-opaque.
There is a broader moral here. The overt data is ambiguous and leaves multiple analyses open: some of these
analyses may be opaque but not all are. Suppose learners overcome the inherent ambiguity in the data and
converge on a shared solution. If they do, that’s because they understand the nature of the process (e.g. why
backing is triggered V’s and not C’s) in a way that makes it possible to predict the full list of trigger and context
segments from some essential property of it, which makes it unnecessary to rely on fragmentary data. Then, if
we establish that learners do this, that’s what linguists should strive for too.
In this case the question is whether they analyse the alternations they’re exposed to in terms of classificatory
features or in terms of articulatory gestures: what quality should a segment have in order to turn ç to x?
5. Â -lenition and ç-backing: basic OT analysis
1
No ç occurs after non-coronal or non-sonorant C’s in German within a single morpheme. The rule of ç-backing is blocked
across morpheme boundaries (hence Kuh-chen ‘cow-DIM’ is [ku:ç´n], not [ku:x´n]): this makes it impossible to determine what
allophone of the dorsal fricative will occur after non-coronal or non-sonorant C’s.
24.962
Page 2
5/23/05
*[+cons, rhotic] in coda >> *x >> Ident [±cons]
*[-back, -cons][+cont, +cons, +back] >> *å8 >> Ident [±back]
Ignoring RoB issues, an analysis is possible for opaque items like fürchten, as follows:
fyÂçt´n
{Ident back
*[+cons,rhotic] *[-back, -cons]
ØIdent
in coda
[+cont, +cons,+back] [±cons]
a { fyÂçt´n
*!
b
F fyå8çt´n
*
*
c fyå8xt´n
*!
*
{Ident [±back]
*[+cons,rhotic] in coda
*[-back, -cons][+cont, +cons,+back]
*å8
*x
ØIdent [±cons]
Ident [±back]
Now we consider RoB and observe that ØIdent [±cons] works only if all instances of R are [+cons] in UR. That
can’t be under RoB: the rhotic emerges as [+cons] and [-cons] entirely as a function of context. Therefore the UR of
German [R] should be free to contain either [-cons] or [+cons] values: the right constraint hierarchy will distribute
these correctly in context. But if [R] is allowed to be either [+cons] or [-cons] in UR, a lexical entry like /fyå8çt´n /,
with coda /å8/ cannot be excluded. Such an item will predict the wrong [x] allophone:
fyå8çt´n
a
b
c
{Ident
back
L fyå8çt´n
{F fyå8xt´n
fyRçt´n
*[+cons,rhotic]
in coda
*[-back, -cons]
[+cont, +cons,+back]
*!
*!
ØIdent
[±cons]
*
One can also derive fyå8xt´n from UR /fyå8xt´n/
/fyå8xt´n/
a
b
c
{Ident
back
L fyå8çt´n
{F fyå8xt´n
fyRçt´n
*[+cons,rhotic]
in coda
*[-back, -cons]
[+cont, +cons,+back]
*!
*!
ØIdent
[±cons]
*
6. One conservative solution: grant the modest increase in serialism involved in setting up one intermediate
derivational step between the Rich Base and the surface form.
Rich Base (universal)
UR’s of the language
H1
SR’s of the language
H2
H1 and H2 are distinct constraint hierarchies. How they differ is open to argument but one option, if we seek to
establish an UR alphabet of segments, is to eliminate from H1 all but rankings of the form M >>F where M is a
context free markedness constraint and F is the corresponding faithfulness. So H1 looks like this in German
(act:
*x
*å8
Ident [±back]
ØIdent [±cons]
24.962
Page 3
5/23/05
Then H2 contains all the rankings established in (5). These work now because the pass through H1 has
eliminated some forbidden segments, like x and å8, from the input to H2.
The option considered by Ito and Mester has the same flavor but allows H-switches at several other junctures
and a less constrained difference between H1 and H2. Likewise for possibilities in Stratal OT.